Distinguishing causation from genetic correlation in a Mendelian randomization framework

The kidney is involved in a number of physiological processes, including fluid and electrolyte balance. Kidney function has been associated with obstructive airways disease, however the nature of this relationship has been difficult to unpick¹. Specifically, it is unclear whether kidney function affects airways disease, vice versa, or whether the association is instead attributable to confounding factors. Insight into the nature of this relationship is of clinical relevance, as the identification of renal impairment as a causal risk factor for obstructive lung disease would offer mechanistic insight and therapeutic opportunities. However, conventional observational studies exploring this have been unable to infer causation, due to the possibility of unmeasured or unknown environmental confounding and reverse causation².

Park et al.³ comprehensively explored the relationship between kidney function and obstructive airways disease by employing the Mendelian randomization paradigm⁴. Using genetic variants as instruments to investigate the effect of altering renal function, the authors were able to investigate potential effects on forced expiratory volume (1 second)-to-forced vital capacity ratio (FEV1/FVC) and risk of obstructive lung disease. The increasing availability of large-scale genetic association data through collaborations such as CKDGen⁵ and biobanks such as UK Biobank⁶ has made such exploration through Mendelian randomization experiments increasingly feasible⁷. This is coupled with continuous methodological developments that strengthen the causal inference afforded by such studies⁸.

In the study by Park et al. the exhaustive framework of analyses convincingly distinguished an association of genetically proxied renal function with obstructive lung disease that is driven by a causal effect of renal function from a genetic correlation between these traits with a different interpretation. Genetic correlation, meaning that an exposure and outcome have shared genetic predictors⁹, can occur for several reasons, including a causal effect of the exposure on the outcome, but also a reverse causal effect of the outcome on the exposure, or the association of genetic variants with a confounder that affects both the exposure and outcome. A number of complementary approaches were performed by the authors to differentiate between these scenarios.

By leveraging 141 uncorrelated genetic variants that associated with both creatinine-based estimated glomerular filtration rate (eGFR) and cystatin C-based eGFR as proxies for the effect of modifying renal function, Park et al were able to employ a range of Mendelian randomization methods that make distinct assumptions on the nature of genetic confounding^8,10. Each of these approaches provided evidence consistent with a causal effect of eGFR on greater FEV1/FVC and lower risk of obstructive lung disease (Table). The primary inverse-variance weighted Mendelian randomization analysis supported a causal effect of impaired kidney function on airway obstruction, with sensitivity analyses more robust to the inclusion variants that might affect airway obstruction through pathways unrelated to renal function producing similar estimates. The multivariable Mendelian randomization analysis that adjusted for genetically predicted hypertension, dyslipidaemia, diabetes, obesity, current smoking and height also supported this, suggesting that the identified association between genetically proxied eGFR and FEV1/FVC was unlikely to be attributable to a causal effect acting via any of these traits¹¹. Looking at urine-albumin-to-creatinine ratio as an alternate measure of renal function, the authors found no strong Mendelian randomization evidence of an effect on obstructive lung disease, supporting that it is specifically kidney function as measured by eGFR that affects airway obstruction.

Table. Complementary insight offered by the distinct Mendelian randomization analyses to distinguish a causal effect of the exposure from a genetic correlation with an alternative interpretation.

Analysis	Objective
Primary analysis (inverse-variance weighted method)	Explore primary evidence for causation
Robust methods (including MR-Egger, weighted median, MR-PRESSO)	Check that results are not driven by a minority of variants that are invalid instruments
Multivariable Mendelian randomization	Check that results are not driven by an alternative risk factor
Urine-albumin-to-creatinine ratio as the exposure, instead of eGFR	Investigate whether it is specifically eGFR that affects airway obstruction
FEV1/FVC genetic associations not adjusted for height or smoking	Check that results are not attributable to collider bias
Bi-directional Mendelian randomization	Explore directionality of effect
Additional related outcomes (including FEV1/FVC < 70%, COPD, and asthma)	Explore for consistent evidence of effect across obstructive airway outcomes
Non-linear Mendelian randomization	Explore shape of the causal relationship

Park et al also acknowledged the adjustment of the FEV1/FVC genome-wide association study for the height and smoking status of participants¹², which has the potential to induce bias in consequent Mendelian randomization analyses¹³. To explore this possibility, the authors repeated the multivariable Mendelian randomization analysis using FEV1/FVC genetic association estimates that did not adjust for height and smoking status, and again achieved similar findings to further strengthen their conclusions.

Bi-directional Mendelian randomization analyses were also performed to investigate whether genetic predictors of obstructive lung disease were associated with kidney function traits. The null findings in these analyses support the notion that the association between FEV1/FVC and eGFR is predominantly attributable to a causal effect of eGFR on obstructive airways disease rather than vice versa.

Considering FEV1/FVC < 70% and chronic obstructive pulmonary disease (COPD) as binary measures of airway obstruction, Park et al further strengthened evidence of a causal effect of eGFR by demonstrating consistency of results across these related outcomes. Analyses exploring asthma as an outcome were able to offer further insight, as Mendelian randomization evidence of an effect of eGFR was only generated when considering late onset asthma, which is more likely to share pathophysiology with COPD¹⁴.

Perhaps the most informative and convincing analysis was non-linear Mendelian randomization of eGFR on FEV1/FEC. The approach taken requires individual participant data, and stratifies the population by their level of the residual exposure (which is the level of the exposure after subtracting the effect of the genetic variants included in the analysis)¹⁵. In their current study, Park et al stratified the UK Biobank population by residual eGFR and then performed Mendelian randomization analyses in each stratum separately. This produced striking evidence that the effect of eGFR on airway obstruction is non-linear, becoming most apparent when eGFR falls lower than approximately 60 mL/min/1.73 m². As it is eGFR levels below this threshold that typically result in clinical consequences, this provides plausibility to the analyses by supporting that it is only biologically relevant changes in eGFR that affect risk of obstructive airway disease.

While the analyses of Park et al offer considerable insight into the relationship between kidney function and obstructive lung disease, they were not intended to pin down the specific underlying mechanism. The authors offer some possible explanations for their findings, including that higher volume status may disrupt alveolar and lung compliance and consequently cause obstructive lung disease^16,17. Further study is therefore required to explain the Mendelian randomization findings before clinical translation should be pursued. Clinical trials investigating the effect of improving renal function or volume status on the prevention and treatment of obstructive lung disease represent the gold-standard for investigating clinical efficacy. Of note, sodium-glucose cotransporter 2 inhibitors have been shown to improve kidney function and cause volume depletion¹⁸, thus offering a potential pharmacological strategy by which kidney function may be targeted for treating obstructive lung disease.

In summary, Park et al performed a series of elegant Mendelian randomization analyses to comprehensively and systematically unpick the association between kidney function and obstructive airway disease. They generated evidence supporting that impaired kidney function, as measured by eGFR, increases risk of obstructive lung disease, and further that this effect is most apparent once eGFR falls below 60 mL/min/1.73 m². Further study is now warranted to establish the underlying mechanism explaining the findings of these genetic analyses, before transitioning to clinical trials that can inform on the potential translation of this insight towards patient benefit. This study provides an excellent example of how Mendelian randomization analyses with a complex polygenic exposure can distinguish between a causal effect of the exposure versus a genetic correlation arising without the exposure having a causal effect.